From: Tobias DiPasquale <anany@...>
Date: 2001-12-08T13:21:11+09:00
Subject: [ruby-talk:27894] Re: App server for Ruby?

Todd Gillespie wrote

> Does anyone actually have a necessary and complete definition of
> 'application server' that is not marketroid?
> 
> This is a real question -- I have heard this thrown around in so many
> places that I don't even know what it means anymore.  If a person
> wishes to write a Ruby application server, what would that actually
> mean?
> 

I cannot tell you what an app server is for sure. Dictionary.com didn't 
have a definition. ;-) As far as I know, from what they tell me in 
class, it could be a web server, a telco switching station or a 
remote-control car. It's sort of like legacy code: legacy code is any 
code you or someone else has written up to five seconds ago.

However, RubiCTM (http://rubictm.linuxninja.net/) is not an "app 
server". Mostly I hear this term in reference to dynamic Web content 
generating something-or-others. My project is a "component transaction 
monitor", which has something of a more tight definition. A CTM is a 
combination of three technologies: message-oriented middleware (the 
ability to have components respond to asynchronous messages), a 
transaction monitor (which really means a DISTRIBUTED transaction 
monitor, since these days, most databases have transactional 
capabilities inherent) and a component broker.

The component broker is usually what people don't get, and it is rather 
ethereal when you think about it. But my understanding of a component is 
that it is an object, or group of objects, that relies on each other and 
the underlying system (broker) to complete some basic task in the system 
as a whole, whether that system be the application you are working on, 
or the whole Internet itself. The OMG has pages and pages of material 
describing components, but that's basically what it is in a nutshell. 
Thus, a component broker's job is to provide certain basic services to 
components, such as persistence, security (mostly access control, via 
ACLs) and interaction with the transaction monitor and MOM pieces of the 
CTM, not to mention the actual ability to be accessed from remote. 
(whether it be via RPC, RMI, XML-RPC, IIOP, SOAP, whatever...)

Some of the major "app servers" you are no doubt hearing about are 
actually CTMs in disguise, ported and updated from the salad days of 
CORBA. For instance, WebSphere is a Java-powered reincarnation of their 
CORBA-powered ComponentBroker, and is a full-fledged CTM (having all 
three of the pieces I mentioned above). The market leader, BEA's 
WebLogic is a similar story, being translated to use EJB and JSP 
technology from their original purchase, a CTM called M3 (which was 
powered by CORBA). Most of the major companies can afford to put the 
development time and effort into creating a full-fledged CTM, so they 
do, because features sell. However, something like Zope or Tomcat is 
severly limited when compared to these big boys, and cannot compete in 
the feature area, even though they might suit your needs perfectly fine! 
(Zope/Tomcat are cool, I use them both) However, the marketing is what 
is really making it hard to tell what these things are. Chances are, if 
it has MOM-capability, ORB-functionality and is TPM aware, it's a CTM. 
If not, it's a dynamic web server technology of some sort. That's 
basically what I can figure... Hope that helps.

-- 
Tobias DiPasquale
Solaris System Administrator
Electrical and Computer Engineering Dept.
Villanova University
mailto: anany@ece.vill.edu
tel: 610-519-5109